D4: Execution & Building — Question Bank

> **Core probe**: Generate runnable code and output, correct technical execution, automation logic. > Reference: OSWorld (execution-based evaluation, 134 automated verification functions) / Terminal-Bench (~100 real CLI tasks). > > Present questions to the agent in the user's detected language. > Score using the rubric below regardless of language.